I note that the man page for humanize_number didn't keep up with
sommerfeld's addition of the 'divisor' argument.  I have changes
to update that man page, but I have another nit that I'd like to
either document or fix...

If you have:
        char    buffer[8];

        format_bytes(buffer, sizeof(buffer), 4608ULL);
        format_bytes(buffer, sizeof(buffer), 128 * 1024ULL);

...you'll get:
	4608\0
	128 KB\0

in both cases.  Also in both cases, format_bytes will return 6 because it
returns the value that humanize_number did.  Humanize_number returned the
first buffer as

	4608 B\0

and format_bytes truncated it.  These functions are defined to return

	the number of characters stored in buffer (excluding the
	terminating NUL) upon success, or -1 upon failure.

I'm inclined to say that they should return the number of characters
stored in buffer prior to the terminating NUL.  This would mean a
simple adjustment to the return value as well as to the buffer if
the " B" is truncated.  However, we lose a little bit of (IMO
irrelevant) information if we do that (i.e., that a couple extra
bytes of buffer beyond the current terminating NUL were touched).

So, does anyone have a problem with changing the return value of
format_bytes in this way?

-allen